The Advisory Board OS.

It’s a different job. A chat gives you prose from one model that agrees easily and forgets you. Here a named expert takes a position (with a Skeptic seated against it), the answer arrives as a deliverable — plan, forecast, decision — with sources and the model named, and your world carries into every next session.

1 credit = $0.01 USD. A consult typically runs 5–40 credits and a full board 55–150 (live 60-day percentiles), so the free 150/month covers real work — and every session shows its cost before you run it.

The frontier set — OpenAI GPT-5, Anthropic Claude, Google Gemini, xAI Grok, with Perplexity Sonar grounding live-web research. Different experts run on different models (that’s the point: rival labs argue), and every answer names the model it ran on.

Yes — with your approval, never without it. The board proposes a write, you preview exactly what it will do, and only then does it execute — with a receipt linking the created resource and a 30-second undo on every reversible write. A risky write gets held a second time, with a seated Skeptic’s objection quoted in front of it. Creating real Notion pages, plans and databases is live today; more write surfaces open as they’re proven.

It becomes your World Model — a page you can read, correct line by line, or wipe. Connected tools are read with permission, every write needs your approval, and undo holds for 30 seconds.

Yes — and it’s built to say so. Conviction is capped at 90 by design, unverifiable claims are logged as such, and dissent stays on the page. Confidence theater is the failure mode we optimized against.